Helsinki University of Technology. Faculty of Engineering and Architecture. Department of Applied Mechanics. Series AM Teknillinen korkeakoulu. Insinööritieteiden ja arkkitehtuurin tiedekunta. Sovelletun mekaniikan laitos. Sarja AM Espoo 2008, FINLAND

TKK-AM-6

ACCIDENT PROBABILITIES IN SELECTED AREAS OF THE GULF OF FINLAND

Ylitalo Jutta Hänninen Maria

Kujala Pentti

Helsinki University of Technology Faculty of Engineering and Architecture Department of Applied Mechanics Teknillinen korkeakoulu Insinööritieteiden ja arkkitehtuurin tiedekunta Sovelletun mekaniikan laitos

Distribution: Helsinki University of Technology Department of Applied Mechanics P.O. Box 4100 FIN-02015 TKK Tel. +358-9-451 3501 Fax +358-9-451 4173  Jutta Ylitalo, Maria Hänninen, Pentti Kujala ISBN 978-951-22-9725-2 ISBN 978-951-22-9726-9 (PDF) ISSN 1797-609X ISSN 1797-6111 (PDF)

Printed in Multiprint Espoo 2008, FINLAND

HELSINKI UNIVERSITY OF TECHNOLOGY ABSTRACT PO Box 1000, FI - 02015 TKK 17.12.2008 http://www.tkk.fi/ Faculty Department Engineering and Architecture Applied Mechanics Author(s) Ylitalo, Jutta; Hänninen, Maria; Kujala, Pentti Title Accident Probabilities in Selected Areas of the Gulf of Finland Abstract In this study, ship-ship collision and grounding probabilities are estimated for several crossing areas and narrow passages and for one grounding location in the Gulf of Finland. The estimates are calculated with AIS-data from July 2006. In addition, the collision probability for the crossing area between Helsinki and Tallinn is estimated for winter month March 2006. Estimates of crossing collision probabilities in 2015 are calculated and compared to the probabilities of 2006. For the crossing area between Helsinki and Tallinn, some expected oil spill sizes and their probabilities are also estimated. Collision probabilities are estimated as a product of geometrical and causation probabilities. The geometrical probabilities are described as the expected numbers of collision candidates, which are estimated with a model derived from literature. Causation probabilities for the collisions are estimated with a Bayesian network model. Grounding probabilities are estimated with three models and with causation probability values derived from literature. For the collision probability estimates in the crossing areas in 2015, the number of tankers heading to and from Russia is assumed to double compared to the year 2006, and for the number of cargo ships, coefficient of 1.5 is applied for the increase of traffic. Based on collision probabilities, oil leak probabilities and expected oil spill sizes are estimated with models and assumptions derived from literature. The results showed that collision probability is highest in the crossing area between Helsinki and Tallinn, where the average time period between accidents for summer traffic is estimated to be five years. In 2015 the average time period is estimated to be about three years. The average causation probability value for the crossing areas is estimated to be 2.7 ∙ 10-4 and 1.0 ∙ 10-5 for head-on encounters. Average time period between cargo oil spills in the crossing between Helsinki and Tallinn in 2006 is estimated to be 126 years, the average spill size being 3200 tons. Bunker oil spill of 330 ton average size is estimated to occur once in 54 years. For groundings, there is a lot of uncertainty related to many of the parameters of applied models, and none of the models estimates the probability of grounding in a sound way in the studied location.

Keywords (and classification) marine traffic safety, collision probability, the Gulf of Finland, Bayesian networks Place Month - Year Language Number of pages Espoo, Finland December -2008 English 49 ISBN (printed) ISBN (electronic) ISSN (printed) ISSN (electronic) 978-951-22-9725-2 978-951-9726-9 1797-609X 1797-6111 Serial name Serial number or report code Series AM TKK-AM-6 Distribution of the printed publication Helsinki University of Technology, Department of Applied Mechanics, P.O. Box 4100, FIN-02015 TKK

5

TABLE OF CONTENTS PREFACE .............................................................................................................................. 7 1

INTRODUCTION ......................................................................................................... 8 1.1 Objective ................................................................................................................ 8 1.2 Estimation of collision and grounding probabilities .............................................. 8 1.3 Bayesian networks ................................................................................................. 9 1.4 Automatic Identification System ......................................................................... 10 1.5 Limitations and report structure ........................................................................... 11

2

TRAFFIC PROPERTIES ............................................................................................ 13 2.1 Overview of traffic in the Gulf Finland ............................................................... 13 2.2 Traffic properties in the studied locations ........................................................... 14 2.2.1 Introduction .................................................................................................. 14 2.2.2 Crossing area between Helsinki and Tallinn ............................................... 19 2.2.3 Crossing in front of Tallinn ......................................................................... 20 2.2.4 Merging traffic from Kotka to the main route of the Gulf of Finland ......... 21 2.2.5 Merging of the waterway to Sköldvik and the main route of the Gulf of Finland… ..................................................................................................................... 21 2.2.6 Merging of two waterways near Sommers .................................................. 22 2.2.7 Merging of lanes from Primorsk and St. Petersburg ................................... 23 2.2.8 Waterway to St. Petersburg ......................................................................... 23 2.2.9 Waterway to Kotka ...................................................................................... 24 2.2.10 Waterway to Vyborg .................................................................................... 24 2.2.11 Grounding location near Sköldvik ............................................................... 24

3

MODELS ..................................................................................................................... 25 3.1 Ship-ship collision probability ............................................................................. 25 3.1.1 Geometrical probability models for crossing and head-on encounters ........ 25 3.1.2 Causation probability model ........................................................................ 27 3.2 Grounding probability.......................................................................................... 29 3.2.1 Pedersen‟s model ......................................................................................... 29 3.2.2 Simonsen‟s model ........................................................................................ 31 3.2.3 The model by Fowler and Sørgård .............................................................. 31 3.2.4 Causation probability for grounding ............................................................ 32 3.3 Probability and magnitude estimation of oil spills .............................................. 32

4

DATA .......................................................................................................................... 33 4.1 AIS-data ............................................................................................................... 33 4.2 Other input data sources ...................................................................................... 34

5

RESULTS .................................................................................................................... 36 5.1 Ship-ship collision probabilities .......................................................................... 36 5.1.1 The number of collision candidates ............................................................. 36 5.1.2 The causation probabilities .......................................................................... 36 5.1.3 Estimated collision probabilities and return periods in studied locations ... 36 5.2 Grounding probability.......................................................................................... 37 5.3 Change in collision probabilities in the future ..................................................... 38 5.4 Estimates of oil spill sizes in the crossing area between Helsinki and Tallinn ... 39

6

SENSITIVITY ANALYSIS ........................................................................................ 40 6.1 Example of ship-ship collisions: the crossing between Helsinki and Tallinn ..... 40

6.2 7

Groundings ........................................................................................................... 41

DISCUSSION .............................................................................................................. 42

REFERENCES..................................................................................................................... 46

7

PREFACE This report has been written within SAFGOF-project, which is a multidisciplinary project conducted in Kotka Research Maritime Centre by Universities of Helsinki and Turku, Helsinki University of Technology and Kymenlaakso University of Applied Sciences. In the SAFGOF- project, the accident risks of marine traffic in the Gulf of Finland are estimated in the current traffic situation and in the year 2015. Also, the direct environmental effects and the risk of environmental accidents can be evaluated. Finally, the effects of national and international legislation and other management actions are modeled, to produce advice and support to governmental decision makers. The aim of this study, conducted by Kotka Maritime Research Centre and Helsinki University of Technology, Department of Applied Mechanics, is to estimate ship-ship collision and grounding probabilities in several crossing areas and narrow passages and in one example grounding location in the Gulf of Finland. For funding of the project, the authors wish to thank European Union, European regional development fund, Regional Council of Kymenlaakso, City of Kotka, Kotka-Hamina regional development company Cursor Ltd., Kotka Maritime Research Association Merikotka, and Kotka Maritime Research Center Corporate Group.

In Kotka, 17.12.2008 Authors

8

1 INTRODUCTION 1.1 Objective The Gulf of Finland is a sensitive geographical area. In 2005, the Baltic Sea, and the Gulf of Finland as a part of it, was categorized as a Particularly Sensitive Sea Area (PSSA) by International Maritime Organization (IMO). The gulf is approximately 400 kilometers long and from 58 to 135 kilometers wide. The average depth of the gulf is only 37 meters and the eastern part is even shallower. Especially the Finnish coast is full of islands. Marine traffic is continuously increasing in the Gulf of Finland. Especially the increasing number of oil tankers is raising concern in coastal countries. Russia is building new oil terminals, and the annual oil transports via the Gulf of Finland are estimated to increase even up to 250 millions of tons by 2015. The risk of an oil accident in the Gulf of Finland is significant. An oil disaster would be devastating for its vulnerable nature. /1, 2, 3, 4/ Authorities are aware of the increasing amount of maritime transport and thus rising risk level. The safety of marine transport is already improved in many ways. In the Gulf of Finland, traffic separation schemes (TSSs) were introduced. In July 2004, the Gulf of Finland mandatory Ship Reporting System (GOFREP) went into operation. It covers the entire sea area in the gulf. Finland, Estonia and Russia manage the system through co-operation. All vessels of over 300 GT are obligated to report before entering the Gulf of Finland or leaving a port in it. The system was launched to improve sea safety and especially reduce the risk of ship collisions. GOFREP operators inform crews about issues affecting the safety or flow of the traffic. They also monitor that regulations are followed. The system relies on the use of radar, AIS (Automatic Identification System) and camera systems. /5/ Several reports concerning grounding and collision risks in the Gulf of Finland have been made, e.g., /6, 7, 8/, but the overall risks of maritime transportation remain to be estimated. It is important to know the present risks in order to make decisions about necessary risk control options and to be sufficiently prepared for possible oil and other accidents. Current risk level also has to be known in estimating the future change in risk level due to increasing traffic. In this study, the existing models for estimating the ship-ship collision and grounding probabilities were applied to several crossing areas and narrow passages and one example grounding location. The estimates were calculated for summer traffic, but for the crossing area between Helsinki and Tallinn, collision probability was estimated for wintertime also. Estimates of some collision probabilities in 2015 were calculated and compared to the probabilities of 2006. For the crossing area between Helsinki and Tallinn, some expected oil spill sizes and their probabilities were calculated.

1.2 Estimation of collision and grounding probabilities In Probabilistic Risk Analysis (PRA), risk is defined as a product of the probability of an unwanted event to happen and the magnitude of its consequences: risk

the probability of an event the consequences of the event

9 The probability is often described as the number of events per time unit, for example the number of collisions per year. The costs describing the magnitude of the consequences might be for example lost human lives in a year or the cost of cleaning oil spills in a year. The objective of risk analysis is to find out what might happen, how probable it is and what are the consequences. /9/ Probabilities of collisions and groundings in marine traffic have often been modeled based on the approach of Fujii et al. /10, 11/ and Macduff /12/. In the approach the probability of a collision or grounding is calculated as P

PG

PC

(1)

, where PG is the geometrical probability and PC is the so-called causation probability The geometrical probability denotes the probability of a ship being on a collision or grounding course. This can be also described using the so-called number of collision/grounding candidates Na. Na denotes the number of ships that would collide or run aground, if no aversive maneuvers are made. This depends on the properties of ship traffic such as traffic distribution on the studied waterway and ship sizes and speeds. For calculating the geometrical probability, there exist few models. These models and their applications to marine traffic risk assessments have been reviewed in /13/. In the earliest studies (e.g., /11, 12/) the traffic was assumed to be distributed evenly on the waterway. To get more realistic distributions, Automatic Identification System (AIS) data gathered from ship traffic in the area can be utilized. The causation probability denotes the probability of failing to avoid the accident while the ship is being on a collision/grounding course, i.e., the probability of not making an evasive maneuver. An accident candidate may result in an accident for example because of a technical fault or human error. Causation probability quantifies the proportion of cases when an accident candidate ends up grounding or colliding with another vessel. Traditionally the causation probability has been estimated based on the difference in calculated geometrical probability, which solely predicts too many accidents, and statistics-based accident frequency (e.g., /11, 12/). In more modern models the value for causation probability has been estimated by applying risk analysis tools such as fault-trees, e.g., in /14/, or utilizing Bayesian networks.

1.3 Bayesian networks Bayesian networks are directed acyclic graphs that consist of nodes representing variables and arcs representing the dependencies between variables. Each variable has a finite set of mutually exclusive states. For each variable A with parent nodes B1,…, Bn there exist a conditional probability table P(A | B1, …, Bn). If variable A has no parents it is linked to unconditional probability P(A). /15/

10 Bayesian networks are used to get estimates of certainties or occurrence probabilities of events that cannot or are too costly to be observed directly. For identifying the relevant nodes and the dependencies between nodes, and constructing the node probability tables, both hard data and expert opinions can be used and mixed. When constructing a network, the structure of the network may be known for example by a domain expert opinion but the probability values of states of nodes may be unknown, or there may not be expert judgment available and then the network structure may also be unknown. If there is some data available, the network structure and the parameters could be learnt from it with maximum likelihood estimation, maximum a posteriori estimation or using EM-algorithm, depending on the completeness of the data. /15/ Bayesian networks have been applied in several fields, including risk analysis of military vehicles /16/, modeling the operational accident causation in railway industry /17/, and the reliability of search and rescue operations /18/. They have been used in modeling nuclear power plant operators‟ situation assessment /19/. In Aviation System Risk Model (ASRM) presented by Luxhøj /20/, human factors in aviation accidents were assessed using Bayesian networks and HFACS human error framework. In 2006, utilization of Bayesian network at step 3 of Formal Safety Assessment was suggested in a document /21/ submitted by the Japan body of maritime safety to the IMO Maritime Safety Committee. Bayesian networks have been applied for modeling human factors in marine traffic, and the existing models have been reviewed in /22/.

1.4 Automatic Identification System Automatic Identification System (AIS) provides means for ships to electronically exchange information. The information is also transmitted to Vessel Traffic Services (VTS) where authorities may observe traffic. AIS operates primarily on two dedicated radio channels but it is capable of being switched to alternate channels. AIS data includes static, dynamic and voyage-related information (table 1). Static information is entered on installation of the system and normally needs not to be changed after that. Dynamic information is automatically got from the ship sensors connected to AIS and only „Navigational status‟ needs to be manually changed. Voyage-related information is manually entered and updated. Also short safety-related messages may be send via AIS. /23, 24/ By the end of 2004, AIS had to be fitted to /25/ -

all ships of 300 gross tonnage and upwards engaged on international voyages

-

cargo ships of 500 gross tonnage and upwards not engaged on international voyages

-

all passenger ships irrespective of size

Some vessel types, for example warships and naval auxiliaries, do not have to carry AIS /25/. The report rate of different dynamic AIS information is presented in table 2.

11 Table 1. Different types of AIS information Static information

Dynamic information

MMSI (Maritime Mobile Service Identity)

Ship‟s position with accuracy indication and integrity status Position Time stamp in Coordinated Universal Time

Call sign and name

Voyage-related information Ship‟s draught Hazardous cargo (type)

IMO number

Course over ground (COG)

Destination and Estimated Time of Arrival (ETA)

Length and beam

Speed over ground (SOG)

Route plan (waypoints)

Type of ship

Heading

Location of position-fixing antenna

Navigational status (i.e. underway by engines or at anchor) Rate of turn (ROT)

Table 2. Report rate of dynamic AIS information /24/ Dynamic information

General reporting interval

Ship at anchor or berthed, speed 0-3 knots

3 min

Ship at anchor or berthed, speed 3 knots or more

10 s

Underway, speed 0-14 knots

10 s

Underway, speed 0-14 knots, changing course

3 1/3 s

Underway, speed 14-23 knots

6s

Underway, speed 14-23 knots, changing course

2s

Underway, speed > 23 knots

2s

Underway, speed > 23 knots, changing course

2s

1.5 Limitations and report structure This study concentrates on the probability estimation of the unwanted event in risk assessment. More information on consequence estimations of collisions and groundings can be found in /26/ and /27/. The risk of all accident types is not analyzed in this study, only ship-ship collisions and groundings. Other accident types include collisions with a bridge, quay or floating object, fires, explosions, leakings, storm damages and capsizings. In the Gulf of Finland, groundings and ship-ship collisions have been the most frequent marine accident types /28/. From the probability of ship-ship collisions, the probabilities of overtaking collision and intersection collision are not analyzed in this document. Latter may occur if a ship omits to change course at the bend of the waterway and as a result collides with another vessel.

12 This report is organized as follows. The locations for which the accident probabilities have been estimated are described in chapter 2. Chapter 3 describes the applied models for geometrical and causation probability estimations, and for the probability and magnitude of oil spills. In chapter 4 the applied AIS data and other input for the modeling is presented. The results are presented in chapter 5 and sensitivity analysis in chapter 6. Finally the results are discussed in chapter 7.

13

2 TRAFFIC PROPERTIES 2.1 Overview of traffic in the Gulf Finland As an example of the vessel traffic in the Gulf of Finland, figure 1 presents the traffic based on AIS records on the 1st of July 2006. Main part of traffic had been directed from or to the Gulf of Finland. Some ships had still operated only in the gulf, for example passenger ships and high speed crafts between Helsinki and Tallinn.

Figure 1. Movements of ships in the Gulf of Finland based on AIS-data from one day

Based on July 2006 AIS-records, 1666 vessels had entered the Gulf of Finland, and 1687 had left it on the way to west. 61 % of ships had been cargo vessels, 19 % tankers, 16 % passenger ships, and 4 % other ships. Less than 10 % of the tankers had been chemical and gas tankers. The rest of the tankers had been crude oil and oil products tankers. The length distribution of ships grouped by ship type is shown in figure 2. In most cases, the length had been between 100 and 200 meters. The average length of passenger vessels was the longest, 174.31 m. All ships had been less than 300 meters long but there had been several more than 200 m long tankers navigating in the Gulf of Finland at all times. The oil tankers had been in average larger than the chemical and gas tankers. 1000 200

340

1100

8000

2006: Probability of oil spill per year

4.2 ∙ 10-4

5.0 ∙ 10-3

2.5 ∙ 10-3

2015: Probability of oil spill per year

8.1 ∙ 10-4

9.7 ∙ 10-3

5.2 ∙ 10-3

Average spill size (tons)

Table 29. Average bunker oil spill sizes and probabilities due to a collision Length of ship (m)

< 100

100-200

> 200

130

360

950

2006: Probability of oil spill per year

6.2 ∙ 10-3

1.3 ∙ 10-2

1.3 ∙ 10-3

2015: Probability of oil spill per year

9.5 ∙ 10-3

1.9 ∙ 10-2

1.8 ∙ 10-3

Average bunker spill size (tons)

40

6 SENSITIVITY ANALYSIS 6.1 Example of ship-ship collisions: the crossing between Helsinki and Tallinn The sensitivity of the geometrical collision model was examined in the crossing between Helsinki and Tallinn in July by altering a value of one parameter at a time while keeping the other parameters unchanged. The effects of modifications to parameter values on the number of collision candidates are presented in table 30. From table 30 it can be seen that the change in traffic volume resulted larger increase in the number of collision candidates than the changes in other tested parameters. Table 30. The effects of altering a geometrical collision probability model’s parameter value on resulting number of collision candidates in the crossing between Helsinki and Tallinn Parameter to be changed

Magnitude of change

Resulting number of collision candidates

-

-

61 (no change)

Ship length

+ 10 m

65

Ship width

+5m

63

Meeting angle

+ 10 deg

59

Traffic volume

x 1.5

137

Ship speed

+ 5 knots

45

The Bayesian network model for the causation probability estimation consisted of 56 nodes and 1782 probability values of in total. Most of the probability values were derived from /33/ and /34/ and had been estimated by experts. This feature and the nature of the phenomenon the model was describing suggest that there was a lot of uncertainty related to the model parameters. The sensitivity of the network was tested with Hugin by setting some states‟ probability to one, thus examining situations where there was certain knowledge on some of the parameters‟ states. Testing revealed that the node “Give-way” had a large role on collision probability. If the probability of loss of control was set to zero, there was still a 2.4999 ∙ 10-4 probability of a collision at the crossing between Helsinki and Tallinn. The main reason for this was that the probability table of the node “Give-way” was based on the assumption derived from /33/, that even when the control was not lost and the ship was supposed to give way, there was a 2.0 ∙ 10-4 probability of collision.

41

6.2 Groundings The probability of grounding using different a/L ratio values was already presented in results at chapter 5.2. It is not obvious that the value of a, the average length between position checks by the navigator, depends on the length of the ship. Some other approaches for estimating the parameter could be adopted instead. If a = 50 m, Pedersen‟s model estimated 1.2 ∙ 10-40 groundings per month like July 2006 and Simonsen‟s model estimated 2.5 ∙ 10-15 groundings. If the value of a was set to the distance moved during 180 s, Pedersen‟s model estimated 3.4 ∙ 10-2 groundings in a month or 4.9 in a year and Simonsen‟s model estimated 5.6 ∙ 10-3 groundings per month or 0.80 in a year. With this value for the parameter a, Pedersen‟s model estimated a larger number of possible groundings than Simonsen‟s model. With small values of a, the results of the models differed a lot from each other and from the result of the model of Fowler and Sørgård. When a increased, the results of all three models converged. When value of a was set to the distance moved during 180 s, the result of the model of Fowler and Sørgård model was between the results of Pedersen‟s and Simonsen‟s models. Especially the Pedersen‟s model was very sensitive to the value of a. Hardly even professional navigators can estimate a so exactly that it would not give a large uncertainty to the results. In Simonsen‟s model it was assumed that the event of checking the position of the ship would to be a Poisson process which made its result less sensitive to a compared to Pedersen‟s mode but still not completely insensitive.

42

7 DISCUSSION Collision and grounding probabilities were estimated in certain locations in the Gulf of Finland. The probabilities were estimated as products of geometrical and causation probabilities. Geometrical probabilities were estimated with commonly applied models. Causation probabilities of collisions were modeled with a Bayesian network based on a model derived from literature. Causation probability values that were applied to grounding probability estimation were based on literature. Collision probabilities were also estimated for the year 2015. Finally, the probability and magnitude of oil leak in the crossing between Helsinki and Tallinn were estimated. As a light of collision probability, the most dangerous location of all studied locations is the crossing of Helsinki-Tallinn traffic and the main traffic of the Gulf of Finland heading to and from east: it was estimated that there occurs 1 collision in every 5 years. If a tanker collided with another vessel in one of the crossing waterways, the location would most probably be the crossing between Helsinki and Tallinn. As is shown in figure 4, there are several small merging locations around the area. They also contribute to the collision risk in the area even though they were not taken into calculations in this study. For the head-on situations, the waterway to St. Petersburg had the highest collision probability and thus the shortest mean time between collisions: 72 years. This is a bit surprising because the location is not in general considered as a very high risk area in the Gulf of Finland. In the crossing in front of Tallinn, the collision probability seems to be relatively low. Collisions were not predicted to occur often in the merging of Primorsk and St. Petersburg waterways either, but if one should take place, it would be very likely that a large oil tanker would be involved, which increases the risk of the accident by increasing the seriousness of the consequences. The head-on collision probability in Kotka waterway does not appear particularly alarming. The waterway to Vyborg does not seem a risky one either, unless the traffic volume will increase considerably. Since the collision and grounding accidents are still quite rare in the Gulf of Finland, it is not very sensible to compare the accident probabilities estimated with the models to accident statistics. However, according to accident statistics, there had been one collision in six years near the studied crossing area between Helsinki and Tallinn /28/. Although the representativeness of statistics from six years period can be questioned, the model results in this area seem to be realistic. In literature the applied value of causation probability has been varying between 6.8 ∙ 10-5 – 6.0 ∙ 10-4 for crossing ships and between 2.7 ∙ 10-5 – 6.0 ∙ 10-4 for head-on encounters, depending also on the studied traffic conditions and ship types /12, 7, 31, 14, 47, 48/. If the Pc values estimated with the model are compared to the values from literature, it can be stated that the results were about the same order. The model applied in this study had been based on models in /33/ and /34/, and in /33/ the probability of a collision given critical situation for large passenger ships was estimated to be 8.4 ∙ 10-6. This is somewhat lower than the estimates in this study. This is mainly because of the difference in the definition of critical course: in this study it was assumed that all critical courses will lead to collision if

43 none of the ships will give way, in /33/ it was assumed that in 20 % of the cases where no evasive maneuver is made, the collision is still avoided. The estimate for the value of causation probability is only one outcome of modeling the accident causation process. The acquired structure and causal relationships describing the process of failures related to human and organizational factors are maybe even more important. The applied Bayesian network is not a comprehensive description of the all factors behind collision causation. In the future, a network that is trying to capture the essential parameters and their occurrence probabilities related to the marine traffic in the Gulf of Finland will be constructed based on expert opinion and bridge simulator studied. The network will also include parameters for studying the effects of different traffic scenarios and risk control options on collision probabilities in the Gulf of Finland. Corresponding network will be constructed also for groundings. The number of ships navigating in the studied waterways is the most influencing parameter of the geometrical crossing collision model. Mainly because of this, the collision probability was estimated to grow in the future. On the other hand, marine traffic safety has been improved in the Gulf of Finland in the 21st century. If it is assumed to continue improving in the future, the applied causation probability model‟s probability values should be adjusted accordingly. The number of collision candidates in the crossing between Helsinki and Tallinn during March was only 41 % of the value estimated with summer traffic. The difference reflects the lower number of ship movements in the area in winter: The number of ships that passed the considered area in March is only 62 % of the number in July. The causation probability for winter traffic was almost as small as for summer traffic, and so the resulting collision probability was lower during winter then in the summer. This is contradictory to the statistics, which show that the majority of collisions had happened in ice conditions /28/. The causation probability was not considering the presence of ice, the different nature of winter navigation and its influence on human failure and could be the main reason for the difference in the estimated results and accident statistics. It should also be noted that when ships are navigating in a convoy in an ice channel, the geometrical probability of a collision in the collision model applied in this study would probably be equal to 1.0 all the time. The ship-ship collision model takes into consideration only a meeting situation of two vessels. Particularly in heavily trafficked locations, three or even more ships may approach the area at the same time. A collision is more difficult to avoid when actions of several other vessels need to be observed. The applied accident probability models do not take into account that traffic is not evenly distributed round the clock which might produce “risk peaks”. For example, passenger traffic between Helsinki and Tallinn is more active during the day than at night time. This changes risk level from the estimate given by the model. The mentioned problem could be avoided by discrete-event simulation in which for example timetables of ferries could be taken into account. Dynamic Bayesian networks could then be applied in causation probability estimation.

44 Especially in the crossing between Helsinki and Tallinn, there was large variation in the value of crossing angle. One reason for the large variation is that for example all northbound ships had not been heading to Helsinki but towards more western ports of Finland. The considered crossing is located in the middle of the Gulf and there is plenty of space to choose one‟s route. The situation becomes more complicated if more than two vessels approach the crossing area at the same time. Avoiding other ships may also partly explain also the large variety of crossing angles. There was a lot of uncertainty in many of the grounding models‟ parameters as well. For example, since the mechanisms behind human failures are not well known and the probability of a human failure is heavily influenced by environment, universal values for parameters such as the probability of omission to check position of ship are difficult or even meaningless to estimate. In the future, using Bayesian networks as a modeling tool also for the estimation of geometrical probabilities might be useful. The uncertainty in the parameters such as the crossing angle would be included in the model if the parameters would be expressed as probability distributions instead of average values. The results of the three grounding models varied significantly. For the considered area, it is impossible to decide which model would give the most realistic estimate. A lot of accident data would be needed for choosing and validating a model. In the future, a model that is suitable for estimating the grounding probabilities in the Gulf of Finland will be developed. All analysis completed in this document is based on the traffic of only two months, July and March 2006. Thus, the amount of traffic during a month is more or less random. In some of the studied waterways, there were much more traffic to one direction than to another. Presumably it is often due to randomness and not the average situation. A part of the traffic is regular, for example many passenger ships operate on the same schedule for several months. The amount of traffic is largely dependent on season as can be seen if the traffic of July and March are compared. Most of the estimates of this study are calculated with summer traffic and thus the yearly estimates of may not be accurate. The AIS records did not contain perfect information. Ship observations without MMSI number, latitude or longitude had not been stored in the database. Thus, the estimated collision probabilities might be lower than the actual frequencies. There were also some blanks in some fields of observations, such as the length, speed or course over ground. Small vessels do not have to carry AIS but they certainly raise the risks of collision and grounding. A large number of recreational boats and fishing ships navigate in the Gulf of Finland. Especially in summertime there are many pleasure boats crossing the Gulf between Helsinki and Tallinn, which were excluded from the calculations since they do not carry AIS. These boats raise the already high risk of collision in the area as well as grounding risk because sometimes vessels may end up off-waterway when trying to avoid a collision with a pleasure boat. There are more crossing and merging areas in the Gulf of Finland than those considered in this document. It is often impossible to tell exactly where a shipping lane has converged to another waterway. Especially in open sea, a ship may continue its voyage long way parallel

45 to the actual waterway where the majority of vessels are navigating. Such a behavior is difficult to model. However, the most distinctive crossing and merging areas contribute the most to the collision probability of the whole Gulf of Finland.

46

REFERENCES /1/

Sonninen, S., Nuutinen, M. & Rosqvist, T. Development Process of the Gulf of Finland Mandatory Ship Reporting System. Reflections on the Methods. VTT, Espoo 2006. VTT Publications 614. 120 p.

/2/

Hietala, M. Oil transportation in the Gulf of Finland. Finnish Environment Institute. 29.6.2006 (online) [cited 22.10.2008]. Available in pdf-format:

/3/

Finnish Environment Institute. Oil transportation in the Gulf of Finland through main oil ports – Oil transportation in years 1995-2005 and estimated development by year 2015. 19.2.2007 (online) [cited 16.7.2008]. Available in pdf-format:

/4/

International Maritime Organization. Revised Guidelines for the Identification and Designation of Particularly Sensitive Sea Areas (PSSAs). Resolution A.982(24). 2005.

/5/

Finnish Maritime Administration. GOFREP – Gulf Of Finland Reporting. n.d. (online) [cited 12.8.2008]. Available in www-format:

/6/

Hänninen, S., Nyman, T., Rytkönen, J., Sonninen, S., Jalonen, R., Palonen, A. & Riska, K. Risks of maritime transport in the Gulf of Finland. Preliminary study (In Finnish). VTT Tuotteet ja tuotanto, Espoo, 2002. BVAL34-021198. 106 p.

/7/

Rosqvist, T., Nyman, T., Sonninen, S. & Tuominen, R. The implementation of the VTMIS system for the Gulf of Finland - a FSA study. RINA International Conference Formal Safety Assessment, London, UK 18.-19.9.2002. London 2002. The Royal Institution of Naval Architects (RINA). p. 151-164.

/8/

Nikula, P. & Tynkkynen, V.-P. Risks in Oil Transportation in the Gulf of Finland: “Not a Question of If – But When”. Aleksanteri Institute, University of Helsinki & Nordregio, Nordic Centre for Spatial Development, 2007. CIVPRO Working Paper 2007:7. 27 p.

/9/

Bedford, T. & Cooke, R. Probabilistic Risk Analysis: Foundations and Methods. Cambridge University Press, Cambridge 2001. 393 p.

/10/

Fujii, Y. & Shiobara, R. The Analysis of Traffic Accidents. Journal of Navigation 24(1971)4, p. 534-543.

/11/

Fujii, Y., Yamanouchi, H. & Mizuki, N. Some Factors Affecting the Frequency of Accidents in Marine Traffic. II - The Probability of Stranding and III - The Effect of Darkness on the Probability of Collision and Stranding. Journal of Navigation 27(1974)2, p. 239-247.

47

/12/

Macduff, T. The probability of vessel collisions. Ocean Industry 1974, p. 144-148.

/13/

Hänninen, M. & Kujala, P. Modeling of Collision and Grounding Risks in Marine Traffic. Literature review (in Finnish). Helsinki University of Technology, Ship Laboratory, Espoo 2007. M-299. 65 p. Available in pdf-format:

/14/

Fowler, T. G. & Sørgård, E. Modeling Ship Transportation Risk. Risk Analysis 20(2000)2, p. 225-244.

/15/

Jensen, F. V. & Nielsen, T. D. Bayesian Networks and Decision Graphs, 2nd Edition. Springer, New York 2007. 447 p.

/16/

Neil, M., Fenton, N., Forey, S. & Harris, R. Using Bayesian belief networks to predict the reliability of military vehicles. IEEE Computing & Control Engineering Journal 12(2001)1, p. 11-20.

/17/

Marsh, W. & Bearfield, G. Using Bayesian Networks to Model Accident Causation in the UK Railway Industry. 2004

/18/

Norrington, L., Quigley, J., Russell, A., & Van der Meer, R. Modelling the reliability of search and rescue operations with Bayesian Belief Networks. Reliability Engineering and System Safety 98(2008)7, p. 940-949.

/19/

Kim, M. C. & Seong, P. H. An analytic model for situation assessment of nuclear power plant operators based on Bayesian inference. Reliability Engineering and System Safety 91(2006)3, p. 270-282.

/20/

Luxhøj, J. T. Probabilistic Causal Analysis for System Safety Risk Assessments in Commercial Air Transport. Workshop on Investigating and Reporting of Incidents and Accidents (IRIA), Williamsburg 16.-19.9.2003. Hampton 2003. NASA. p. 1738.

/21/

International Maritime Organization. Formal Safety Assessment. Consideration on utilization of Bayesian network at step 3 of FSA. Maritime Safety Committee 81st Session. IMO, London 2006. MSC 81/18/1. 4 p.

/22/

Hänninen, M. Analysis of Human and Organizational Factors in Marine Traffic Risk Modeling: Literature review. Helsinki University of Technology, Department of Applied Mechanics, Espoo 2008. TKK-AM-4. 51 p.

/23/

International Maritime Organization. Guidelines for the Onboard Operational Use of Shipborne Automatic Identification System (AIS). IMO, London, 2002. 14 p.

/24/

Finnish Maritime Administration. Automatic Identification System (AIS). n.d. (online) [cited 12.8.2008]. Available in www-format:

48

/25/

International Maritime Organization. IMO www pages. n.d. (online) [cited 18.7.2008]. Available in www-format:

/26/

Pedersen, P. T. Collision and Grounding Mechanics. Proceedings of WEMT'95. Copenhagen 1995. The Danish Society of Naval Architects and Marine Engineers. p. 125-157.

/27/

Simonsen, B. C. Mechanics of Ship Grounding. PhD Thesis. Technical University of Denmark, Department of Naval Architecture and Offshore Engineering. Kgs. Lyngby 1997. 260 p.

/28/

Kujala, P., Hänninen, M., Arola, T. & Ylitalo, J. Analysis of the marine traffic safety in the Gulf of Finland. 2009. Submitted to Reliability Engineering and System Safety.

/29/

Pedersen, P. T. & Zhang, S. Collision Analysis for MS Dextra. SAFER EURORO Spring Meeting, Nantes, France 28.4.1999. Paper No. 2, p. 1-33.

/30/

Germanischer Lloyd. DEXTREMEL Final Technical Report. 2001 (online) [cited 11.8.2008]. Available in pdf-format: .

/31/

Otto, S., Pedersen, P. T., Samuelides, M. & Sames, P. C. Elements of risk analysis for collision and grounding of a RoRo passenger ferry. Marine Structures 15(2002)4, p. 461-474.

/32/

Rambøll. Navigational safety in the Sound between Denmark and Sweden (Øresund); Risk and cost-benefit analysis. Rambøll Danmark A/S, 2006 (online) [cited 5.6.2008]. Available in pdf-format: .

/33/

Det Norske Veritas. Formal Safety Assessment – Large Passenger Ships, ANNEX II. n.d. (online) [cited 30.5.2008]. Available in pdf-format: .

/34/

Det Norske Veritas. Formal Safety Assessment of Electronic Chart Display and Information System (ECDIS). 2006. Technical Report No. 2005-1565, rev. 01. 111 p.

/35/

Hugin Expert A/S. Hugin. n.d. [cited 22.10.2008]. Trial version can be downloaded at: .

/36/

Simonsen, B. C. Mechanics of Ship Grounding. PhD Thesis. Technical University of Denmark, Department of Naval Architecture and Offshore Engineering. Kgs. Lyngby 1997. 260 p.

49

/37/

Fujii, Y. & Mizuki, N. Design of VTS systems for water with bridges. In: Gluver, H. & Olsen, D. (ed.). Ship Collision Analysis: proceedings of the International Symposium on Advances in Ship Collision Analysis. Rotterdam 1998, Balkema. p. 177–90.

/38/

MacDonald A., McGeehan C., Cain M., Beattie J., Holt H., Zhou R. & Farquhar, D. Identification of Marine Environmental High Risk Areas (MEHRA's) in the UK. Department of the Environment, Transport and the Regions, London 1999. ST-87639-MI-1-Rev 01.

/39/

International Maritime Organization. Åland Sea FSA Study. Sub-Committee on Safety of Navigation, 54th session, IMO London 2008.

/40/

Det Norske Veritas. A Simple Model of The Costs of Ship Accidents Rev 3. 2003. 34 p.

/41/

Finnish Meteorological Institute. Weather statistics - humidity and fogs (in Finnish). n.d. (online) [cited 19.11.2008]. Available in www-format:

/42/

Finnish Meteorological Institute. Weather statistics - wind and storms (in Finnish). n.d. (online) [cited 19.11.2008]. Available in www-format:

/43/

Finnish Maritime Administration. FMA www pages. n.d. (online) [cited 18.10.2008]. Available in www-format:

/44/

Central Marine Research and Design Institute (CNIIMF). VTMIS and AIS Network in Russia (the Gulf of Finland). Present State and News. November 2006.

/45/

Finnish Environment Institute. Oil transportation in the Gulf of Finland through main oil ports – Oil transportation in years 1995-2005 and estimated development by year 2015. 2007 (online) [cited on 16.7.2008]. Available in pdf-format: .

/46/

The Institute of Shipping Analysis (in Sweden), BMT Transport Solutions GmbH (in Germany) & Centre for Maritime Studies (in Finland). Baltic Maritime Outlook 2006. Goods flows and maritime infrastructure in the Baltic Sea Region. 2006. 112 p.

/47/

Fujii, Y. Integrated Study on Marine Traffic Accidents. IABSE Colloquium on Ship Collision with Bridges and Offshore Structures, Copenhagen 1983. p. 91-98.

/48/

Karlsson, M., Rasmussen, F. M. & Frisk, L. Verification of ship collision frequency model. In: Gluver, H. & Olsen, D. (eds.). Ship collision analysis: Proceedings of the International Symposium on Advances in Ship Collision Analysis. Rotterdam 1998, Balkema. p. 117-121.